What Is the Storage Effect, Why Should It Occur in Cancers, and How Can It Inform Cancer Therapy?

Home , Storage effect

Utah State University DigitalCommons@USU

Ecology Center Publications Ecology Center

7-29-2020

What Is the Storage Effect, Why Should It Occur in Cancers, and How Can It Inform Cancer Therapy?

Anna K. Miller H. Lee Moffitt Cancer Center and Research Institute

Joel S. Brown H. Lee Moffitt Cancer Center and Research Institute

David Basanta H. Lee Moffitt Cancer Center and Research Institute

Nancy Huntly Utah State University

Follow this and additional works at: https://digitalcommons.usu.edu/eco_pubs

Part of the Ecology and Evolutionary Biology Commons

Recommended Citation Miller, A. K., Brown, J. S., Basanta, D., & Huntly, N. (2020). What Is the Storage Effect, Why Should It Occur in Cancers, and How Can It Inform Cancer Therapy? Cancer Control. https://doi.org/10.1177/ 1073274820941968

This Article is brought to you for free and open access by the Ecology Center at DigitalCommons@USU. It has been accepted for inclusion in Ecology Center Publications by an authorized administrator of DigitalCommons@USU. For more information, please contact [email protected]. Commentary & View

Cancer Control Volume 27: 1-9 ª The Author(s) 2020 What Is the Storage Effect, Why Should It Article reuse guidelines: sagepub.com/journals-permissions Occur in Cancers, and How Can It Inform DOI: 10.1177/1073274820941968 Cancer Therapy? journals.sagepub.com/home/ccx

Anna K. Miller, PhD1 , Joel S. Brown, PhD1, David Basanta, PhD1, and Nancy Huntly, PhD2

Abstract Intratumor heterogeneity is a feature of cancer that is associated with progression, treatment resistance, and recurrence. However, the mechanisms that allow diverse cancer cell lineages to coexist remain poorly understood. The storage effect is a coexistence mechanism that has been proposed to explain the diversity of a variety of ecological communities, including coral reef fish, plankton, and desert annual plants. Three ingredients are required for there to be a storage effect: (1) temporal variability in the environment, (2) buffered population growth, and (3) species-specific environmental responses. In this article, we argue that these conditions are observed in cancers and that it is likely that the storage effect contributes to intratumor diversity. Data that show the temporal variation within the tumor microenvironment are needed to quantify how cancer cells respond to fluctuations in the tumor microenvironment and what impact this has on interactions among cancer cell types. The presence of a storage effect within a patient’s tumors could have a substantial impact on how we understand and treat cancer.

Keywords storage effect, cancer, heterogeneity, coexistence, dormancy, quiescence, fluctuating environments

Received March 5, 2020. Received revised June 12, 2020. Accepted for publication June 19, 2020.

Introduction Long underappreciated in ecology was a mechanism of coexistence now known as the storage effect.7-11 Three ingre- Ecosystems in nature include coexisting species that compete for dients can promote the coexistence of species by the storage space, resources, and safety from predators.1-4 Similarly, tumors effect: (1) temporal variability in environmental conditions that exhibit microenvironmental heterogeneity that contains coexist- include periods favorable and unfavorable for survival and ing cancer cell types that experience variations in nutrients, toxic reproduction (we shall refer to these as good and bad), (2) the metabolites, and diverse types of normal cells.5 In nature, presence of 2 life-history states within each species (one mechanisms of coexistence can explain the diversity of species within a community. Typical mechanisms include food-safety trade-offs (one species is the better competitor, while the other is 1 Department of Integrated Mathematical Oncology, H. Lee Moffitt Cancer better at avoiding predation), diet separation (each species has a Center and Research Institute, Tampa, FL, USA subset of resources on which it is the more successful consumer), 2 Ecology Center & Department of Biology, Utah State University, Logan, UT, habitat selection (each species has a habitat within which it is USA more successful), and competition-colonization trade-offs (one species slowly outcompetes the other at a given spot, while the Corresponding Author: Anna K. Miller, Department of Integrated Mathematical Oncology, H. Lee other is more successful at dispersing to unoccupied spots). Sim- Moffitt Cancer Center and Research Institute, USF Magnolia Drive, Tampa, ilar mechanisms likely explain some of the diversity of cancer FL 33612, USA. 6 cell types within and among a patient’s tumors. Email: [email protected]

Creative Commons Non Commercial CC BY-NC: This article is distributed under the terms of the Creative Commons Attribution-NonCommercial 4.0 License (https://creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/open-access-at-sage). 2 Cancer Control conducive to proliferation during good periods, the other con- infiltration, nutrients, and toxins, the architecture of cells ducive to survival during bad periods), and (3) differences in within and around a tumor microenvironment also may change how species perceive and invest effort to grow, or do not, in temporally. The rate of diffusion of nutrients (such as glucose good and bad periods, reflecting some underlying trade-offs and glutamine) toward a neighborhood of cancer cells and the and adaptation to somewhat different environmental diffusion of metabolites away (such as lactate, free-oxygen conditions. radicals, and pyruvate) may vary temporally as other neighbor- Here, we posit that the storage effect may promote the coex- hoods of cancer cells block or unblock intracellular channels, istence of at least some cancer cell types within the tumors of as immune cells move in or out of an area, and as fibroblasts some or many different cancers. To demonstrate the plausibil- change extracellular matrices and/or the secretion of growth ity of the storage effect, we discuss how, as in ecosystems in factors. Regardless of its source or regularity, coexisting cancer nature, conditions (1) and (2) are universal properties of most cells experience considerable temporal variability in opportu- tumor microenvironments. Next, we discuss why there may be nities and hazards that likely create good and bad periods in trade-offs in the way cancer cells experience good and bad terms of their proliferation and survival rates, setting the poten- periods. We then discuss how the storage effect manifests in tial for the storage effect to promote diversity of cancer cells nature, followed by a simple mathematical model for cancer within tumors. tumor cells, illustrating how the storage effect works. We con- clude with a discussion of how knowledge of the presence of a Proliferative and Nonproliferative storage effect in a patient’s disease might inform therapies. Life-History States in Nature and Cancer Temporal Variation in Nature and the Tumor In response to fluctuations that generate good and bad periods, many organisms have evolved different life-history states: one Microenvironment that is best at exploiting the good times and the other best at Very few, if any, ecological communities experience temporally surviving through the bad times. Many single-celled protists constant environments. Migratory birds escape the harsh winter have a proliferative state that allows the cell to feed, move, and conditions of higher latitudes by migrating to destinations closer proliferate.15 This state, however, may not be able to survive to the equator. Deciduous trees lose their leaves during dry and/ bad periods when their pool of water dries up or the environ- or cold seasons. Year-to-year variation in temperature and pre- ment offers only toxins and no nutrients. Such protists also cipitation may portend droughts or floods and hot spells and cold have an encysted state. This state, although unable to prolifer- snaps. Fire, disease, or pestilence occurs episodically within ate, is highly resistant to desiccation, toxins, and nutrient depri- ecosystems. A beaver damming a stream can raise water tables, vation. This state may be the only way the protist survives drown surrounding terrestrial vegetation, and create a wetland. through bad periods. Having a fraction of the protist population This wetland can revert when the beaver dam breaks or decays. in an encysted state means that some opportunities are lost Virtually, all living organisms experience predictable and during good periods, but survival is ensured during the bad. unpredictable variability in environmental conditions that influ- Baker’s yeast is a classic example. The dry packet of encysted ence growth, reproduction, and survivorship. yeast can survive for years without an opportunity for growth Like natural ecosystems, cancer cells inhabiting a tumor and reproduction. Upon activation and favorable conditions, microenvironment experience both regular and irregular tem- this yeast enters a feeding and proliferative state. Among spe- poral fluctuations. Blood flow changes due to unstable vascu- cies in nature, “quiescence” or “dormancy” refers to a range of lature on time scales of minutes to hours.12 This instability cellular or organismal states characterized by slowed metabo- arises as cancer cells adapt to local hypoxic conditions by lism and relatively high resistance to hardships from their sur- recruiting new blood vessels, which enable the delivery of rounding environment.16-21 nutrients and the removal of toxic metabolites so that cancer In cancer, cells that are reversibly arrested in the G0 phase cells can survive when nutrients are low. However, as cancer of the cell cycle are referred to as quiescent, whereas “cell cells induce the growth of new blood vessels, the vascular dormancy” typically refers to a long-term quiescence.22 Many network becomes more irregular and disorganized, leading to cancer cells respond to hypoxia by becoming quiescent and unpredictable spatial and temporal variations in blood flow.13 upregulating autophagy to survive lower levels of nutrients and Consequently, transient changes in perfusion lead to cycling oxygen.13 Cancer cells may survive harsh environments (eg, hypoxia. Local fluctuations between hypoxia and reoxygena- hypoxia, low pH, toxins) by forming poly-aneuploid cancer tion affect cells adjacent to the poorly perfused blood vessels. cells (also referred to in the literature as polyploid giant cancer This is in contrast to chronic hypoxia, which affects cells far cells), a population of reversibly quiescent cells that form rap- away from vessels due to limited diffusion. Cycling hypoxia is idly (within 72 hours) in response to environmental stress and associated with an increase in cell migration, metastatic poten- later divide asymmetrically to repopulate a tumor with cells of tial, and resistance to treatments compared to chronic normal ploidy that have increased resistance to chemother- hypoxia.14 apy.23,24 In their nonproliferative state, dormant or quiescent While changing patterns of bloodflowlikelycreatethe tumor cells are able to evade treatments that preferentially greatest source of temporal variability in oxygen, pH, immune target highly proliferative cells. After treatment, these cells can Miller et al 3 resume proliferation, which can result in tumor growth and vegetation, cancer cells also would share many environmental relapse. Quiescent states are ubiquitous in cancer and can be needs and so experience good and bad times in part together. associated with metastasis and relapse of cancerous growth.25 However, as with plants that differ in tolerance to low water As such, dormancy and quiescence challenge our ability to availability or high temperatures, cancer cells differ in their treat, control, or moderate cancer. sensitivities to hypoxia, low pH, immune infiltration, the Some readers may see similarities between cancer stem cells absence of growth factors, and the stromal (created by normal (CSCs) and our description of quiescent and dormant cancer cells) architecture of their microenvironment, generating cells. The similarity does manifest when CSC refers to a subset potential for trade-offs that make some periods better for some of heterogeneous cancer cells that exhibit plasticity, drug resis- cell clones and others better for other clones. The heterogeneity tance, and tumor-initiating capability. But, CSC concepts of cancer cells in many features, including their proliferative diverge from our use of quiescence and dormancy when CSCs potential, hormone receptor expression, immunogenicity, sen- are seen as a small population of undifferentiated, self- sitivity to drugs, motility, and angiogenic potential,31 enables renewing cells that give rise to and sustain a population of cancer cells to have population-specific environmental terminally differentiated cells. Once fully differentiated, these responses, an essential component of the storage effect. For cells are viewed as being nonproliferative even as they make up example, in stage 2 invasive breast cancers, cells on the tumor the bulk of the tumor. Cancer stem cells can undergo either edge tend to have an acid-producing invasive, proliferative symmetric or asymmetric cell division, where the former phenotype, whereas those in the hypoxic tumor core have a increases the CSC population and the latter promotes the dif- less proliferative phenotype.32 Tumors also have metabolic ferentiated population. For the storage effect, the proliferative heterogeneity in the form of acid-resistant or glycolytic pheno- state is vulnerable to harsh conditions while the nonprolifera- types.33 Mathematical models of cancer evolution suggest that tive quiescent state is not. Each state can phenotypically shift high nutrient variability (cycling hypoxia) gives a competitive into the other. advantage to cancer cells that have a higher rate of phenotypic In nature, quiescent and resistant states can have a powerful transition, which promotes high levels of phenotypic effect. They can allow populations to survive despite exposure heterogeneity.34 to conditions that limit or preclude population growth. They Cancer cells, and even normal cells, have a wide range of can buffer populations against variation in favorability for behaviors associated with entering and exiting from nonproli- growth on many time scales, including seasonal, annual, and ferative states, which provide the buffering life-history stage multiannual; they can contribute to more diverse communities that enables a storage effect. Cancer cells may remain in these than would exist without them.9,10,26-28 Given how common states for times ranging from the very short (1 to several days) dormancy/quiescence is in cancer, we expect to find similar to the very long (decades, as observed for emergence of meta- ecological effects of these life-history stages for the cancer static cancer growth after years of healthy remission).25,35 In ecosystem. Quiescent or dormant states could maintain diverse fibroblasts, cells move deeper into quiescence following longer cancer phenotypes that may proliferate, coexist, or displace durations of serum starvation or contact inhibition and require a each other over the course of time within the patient. For stronger growth stimulation to exit quiescence compared to instance, disseminated cancer cells that stay dormant, some- cells in shallow quiescence.36 Interactions between cancer cells times for many years, and then reemerge as metabolically and the tumor microenvironment can promote quiescence or active cells that proliferate, often leading to the death of dormancy. For example, bone-lining osteoblasts and endothe- patients. These emergent metastases appear to result from lial cells on stable vasculature secrete factors that keep disse- enhanced cell lineage persistence through dormancy.23-25,29,30 minated tumor cells dormant,37 and adhesion of multiple The movement of cells into and out of dormancy, as the envi- myeloma cells to extracellular matrix components such as ronment around them changes, could affect what clones co- fibronectin leads to quiescence.38 occur in a tumor or in a patient by changing the ways in which The time of exit from quiescence can vary within a clonal cell–cell interactions such as competition play out at the level population, even when cultured in the same growth condi- of populations of different cell types. The presence of quiescent tions.39 Dormant cells awaken due to environmental changes. and dormant, resistant states of cancer cells sets the potential Bone is highly dynamic, undergoing constant remodeling, for the storage effect to promote the diversity of cancer cells which can reawaken dormant cancer cells.40 The growth of within tumors. These possibilities have clear relevance for can- new blood vessels provokes sufficient change to the microen- cer control. vironment to reactivate breast cancer cells.41,42 Age-related inflammation promotes the outgrowth of dormant cells in pancreatic ductal adenocarcinoma liver metastases.43 Poly- Evidence for Population-Specific Behaviors aneuploid cancer cells emerge from quiescence after removal and Trade-Offs in the Way Cancer Cells of chemotherapy, and they do so after a range of quiescence durations (Sarah Amend, PhD, email communication, February Experience Good and Bad Times 2020). Drug resistance is associated with a heterogeneous pop- As in nature, where a year with plentiful but not excessive ulation of nonproliferative cells, including drug-tolerant pers- water would be expected to be generally a good year for isters, therapy-induced senescent cells, hypoxic drug-resistant, 4 Cancer Control and disseminated tumor cells. These populations differ in their proliferative state. A fraction of cells, y, leave the arrested state propensities to remain in or exit the growth-arrested state.22 during each period. Those that enter the proliferative state have a mortality rate that depends on whether it is during a “good” or “bad” period. These periods could represent a microenviron- A Simple Model of the Storage Effect in ment of a tumor where fluctuations in blood flow, hypoxia, Cancer and/or growth factors contribute to times of plenty and times Mathematical models have been useful to understand cancer of famine. Upon emerging from the arrested state, cells suc- biology and to help test what treatment strategies might cessfully survive to proliferate at fraction s. Let sg >> sb be improve patient outcomes over the standard of care. Many these survival rates during good and bad periods, respectively. model types have been used, including game theory mod- The biology here is that if a cell attempts to be proliferative, els,44-51 agent-based models,33,52-57 and Lotka-Volterra-type it may lack key resources even as the cell commits to cell dynamical models that consider cancer clonal cells as interact- division. The consequence is cell mortality that would not have ing populations.58,59 So far, very few of these models consider happened had the cell remained in an arrested state. For exam- variation in the environment of cancer cells in time. It is unsur- ple, providing estrogen to breast cancer cells that otherwise prising that temporal variation in the tumor environment has have no other nutrients can cause mortality, as cells are tricked had little attention in dynamical models. This was long the case into trying to proliferate (Robert Gatenby, MD, email commu- in models of natural ecosystems, and the data available for nication, February 2020). Those cells that survive produce a cancer tend to be static snapshots of a tumor. Radiographic and maximum number of daughter cells, f, that declines with com- histologic images show cancer at a point in time. Although they petition from other cells in a proliferative state. We use a reveal, and provide data for, modeling spatial heterogeneity in Ricker model to describe the competition between proliferative cancer, they do not capture or inform the potential variation in cells for resources. In the absence of competitors, a prolifera- time that small-scale physiological studies indicate is there as tive cell produces its maximum number of daughter cells; with well.12,14,60 competitors, the production of daughter cells declines exponen- Stochastic dynamic models of interacting populations of tially with the density of other proliferative cells. We imagine cancer cells are a straightforward way to incorporate temporal that f can take on a positive value that represents the expected environmental variation and are the class of models that have number of daughter cells that might accrue during a period primarily been used to develop an understanding of the storage sufficiently long to permit 1 or more rounds of cell division. effect. The simple Lotka-Volterra-type population models so At the end of the period, daughter cells enter the pool of cells in far used to study cancer are not sufficient to capture coexis- an arrested state. tence mechanisms such as the storage effect because they do Now let there be 2 cell types that are equal in all ways, not incorporate a nonproliferative life-history state nor do they except that they do not always experience good and bad periods incorporate species-specific demographic responses to a vary- in the same way. Although unlikely that 2 cell types would be 3,61,62 ing environment. Variation in the environment can be identical, this assumption allows us to illustrate coexistence proxied as stochastic variation in demographic parameters, that only manifests because of the storage effect. such as entry to and exit from quiescence and rates of growth, For cancer cell types, it is likely that what is good for one is, reproduction, or survival of cancer cells. in part, good for the other type. However, differences in nutri- To show how coexistence can occur via the storage effect, ent metabolism, sensitivity to growth factors, resistance to we construct a simple discrete-time stochastic model that hypoxia, the ability for immune evasion, and/or physical con- includes 2 species, each with 2 life-history stages (proliferative ditions like pH mean that at times one cell type may experience and quiescent) and stochastic variation in good and bad a bad period while the other a good period. Let q be the prob- periods. ability of a good period, and let 0 r 1 represent the degree to which the 2 cell types experience good and bad periods Illustrating the Storage Effect similarly. The term (1 r) then represents the probability that The plausibility of the storage effect promoting the coexistence the 2 cell types are experiencing the type of period indepen- of different cell types within a tumor microenvironment dently of each other. Thus, the probabilities of both experien- emerges from (1) temporal variability in blood flow, immune cing good periods (pGG), both experiencing bad periods (pBB), infiltration, stromal architecture, and physical conditions and type 1 experiencing good and type 2 experiencing bad (pGB), (2) the widespread occurrence of cell arrest, quiescence, and/or and vice versa (pBG) are given by: 2 dormancy that allows cancer cells to exist in 2 states—prolif- pGG ¼ rq þð1 rÞq 2 erative and nonproliferative. Coexistence can occur if there is a pBB ¼ rð1 qÞþð1 rÞð1 qÞ trade-off in how 2 cell types experience good and bad periods. pBG ¼ pGB ¼ð1 rÞqð1 qÞ: To illustrate how storage effects can work, we propose a simple model that aims to keep assumptions to a minimum As they must, pGG, pBB, pBG, and pGB sum to 1. while containing the essential elements. We imagine that each Let xk(t) be the number of cells of type k in an arrested state cancer cell type has 2 stages: an arrested state and a following time period t, where k ¼ 1 or 2. The 2 species growth Miller et al 5

600

400

200 Number of Cells in Arrested State

0 250 500 750 1000 Time

Total Species 1 Species 2

Figure 1. Invasion of resident population. Parameters: sg ¼ 0.8, sb ¼ 0.1, R ¼ 100, d ¼ 0.05, y ¼ 0.2, f ¼ 5, q ¼ 0.4, r ¼ 0.8, x1(0) ¼ 100, x2(0) ¼ 0, x2(250) ¼ 1. equations take on the following form: experiencing an advantage when rare. Once r < 1 and y <1,the coexistence of cell types by the storage effect will happen, but yðsi1x1ðtÞþsj2x2ðtÞÞ R the strength of this will increase as r approaches 0. The advan- x1ðt þ 1Þ¼ð1 yÞð1 dÞx1ðtÞþysi1x1ðtÞfe tage that a cell type gains when it is rare (thus driving it to yðsi1x1ðtÞþsj2x2ðtÞÞ increase in time at the expense of the other cell type) derives R from the occurrence of periods where the rare species experi- x ðt þ 1Þ¼ð1 yÞð1 dÞx ðtÞþys x ðtÞfe 2 2 i2 2 ences a good and the resident cell type experiences a bad period, where i, j ¼ g or b, and d is the mortality rate of cells in an giving the rare species opportunity to realize the high growth arrested state per period, and R scales the limits to growth by potential of a good period without strong depression by compe- representing some measure of resources available to cells that tition. This is maximized when r ¼ 0 and when q ¼ 0.5 (good and are in a proliferative state. bad periods are equally likely). Furthermore, as a long-term The above model easily generates a coexistence via a storage dynamic equilibrium, the average tumor burden increases as r effect. This can be seen by showing how either species 1 or 2 can goes from 1 to 0; and as q goes from 0 to 1 (Figure 2). invade a resident population of the other and that when together they settle on a dynamic equilibrium but with more or less tem- How Might Consideration of the Storage poral variation in the frequency of the 2 cell types within a microenvironment (Figure 1). When averaged over many Effect Inform Cancer Therapy? microenvironments, the cell type frequencies might appear rel- The existence of the storage effect contributing to diversity in atively stable across time. If a storage effect is operating within a cancer would have consequences for clinical treatments. It is tumor, then there may be surprisingly large fluctuations in cell recognized that intratumor and intertumor heterogeneity should composition at the level of small neighborhoods of cancer cells. inform cancer therapy.63,64 For instance, intratumor heteroge- The storage effect is not possible in the absence of an arrested neity is established as a marker for poor prognosis. Additional state. For y ¼ 1, the 2 cancer cell types exist as a stochastic therapeutic challenges arise if some of the diversity of cancer random walk with neither one having an advantage. An arrested cell types involve storage effect dynamics. The presence of a state buffers a cell type when it experiences a bad period, while storage effect presents a double challenge for therapy. First is the other experiences a good period. Similarly, if r ¼ 1, then a the task of targeting quiescent cells, and second is the challenge storage effect is not possible, as both cell types perceive good of finding drugs effective on each or all of the cancer cell types. and bad years identically. The 2 will simply experience a sto- Quiescence or dormancy now appears to be a major mechan- chastic random walk of population sizes with neither ism by which cancers evade drug therapies and initiate 6 Cancer Control

Average Number of Arrested Cells

1.00 0 0 1 116 299 423 518 593 654 707 752

0 0 1 122 302 429 523 599 660 710 752

0 0 1 124 303 439 533 606 664 713 752 0.75 0 0 1 122 317 443 539 614 671 715 752

0 0 1 131 321 454 548 621 675 719 752 600

r 0.50 0 0 1 134 325 462 558 627 679 722 752 400 0 0 1 136 333 470 564 636 686 724 752 200

0 0 1 140 342 480 572 642 690 727 752 0.25 0 0 1 145 352 485 582 650 697 730 752

0 0 1 142 356 496 590 657 702 732 752

0.00 0 0 1 154 366 506 600 664 707 735 752

0.00 0.25 0.50 0.75 1.00 q

Figure 2. Average number of arrested cells as a function of q and r. Each combination represents an average of 10 simulations. Parameters: sg ¼ 4 0.8, sb ¼ 0.1, R ¼ 100, d ¼ 0.05, y ¼ 0.2, f ¼ 5, x1(0) ¼ 100, x2(0) ¼ 100, number of timesteps ¼ 10 . metastases,23-25 so there is great interest in whether quiescent or continued efficacy may slow, as the remaining cancer clonal dormant cells can be targeted by therapies.65 This article argues populations have lower fractions of cells in a proliferative that we must also consider that there may be many coexisting state. The storage effect, by maintaining the coexistence of clonal lineages of cancer cells that differ in physiologies of their cancer clones that vary in their propensity for remaining quies- nonproliferative and proliferative life-history states. If cancer cell cent or arrested, could exacerbate pseudo-resistance. types coexist via the storage effect, then one is not targeting just a With pseudo-resistance, continuing the same drug regimen single cell type, but a diversity of cell types with potentially would continue to cull cancer cells as they emerge into a pro- differing susceptibilities to therapy. Longitudinally monitoring liferative state. However, the fraction of cancer cells killed with tumors and responses to therapy may require more frequent mag- each dosing would decline, generating the appearance of netic resonance imaging and other imaging techniques as nonin- declining tumor burden, but without entirely eradicating the vasive tools for identifying and tracking different tumor habitats tumor. Prolonging the therapy may increase the likelihood of based on vascularity, hypoxia, necrosis, and so on.66,67 true resistance emerging from the surviving cancer cells and also may induce undue toxicities to the patient. The possibility that pseudo-resistance might reflect a storage effect would sug- Pseudo-Resistance gest novel therapeutic strategies. One approach might be to Heterogeneity in quiescence could obscure detection of resis- maintain the ongoing therapy, but reduce dosage or make it tance to treatment through the phenomenon of pseudo-resis- more intermittent (akin to maintenance therapy for patients in tance, a situation where an otherwise treatment-sensitive remission). Frequent and accurate measures of tumor burden population of cancer cells appears to be resistant. Pseudo- might identify breakpoints in the rate of tumor burden decline, resistance could result from temporal variation that selects which would indicate the presence of different clones with for cell lineages that maintain a high fraction of cells within a different propensities to remain quiescent. A second therapeu- quiescent state. Many cytotoxic chemotherapies target pro- tic approach could be to include in tandem an intervention liferating cells that are actively dividing, and nonproliferative designed to amplify the rate at which cancer cells return from cells may be insensitive to these therapies. Those parts of a quiescence into a proliferative state. It might also be useful to tumor that experience the greatest temporal variability may begin with the first therapy and then add to it, rather than have the greatest fraction of nonproliferative cells. In this case, replace it with, additional therapies; if possible, the second therapy may appear effective in eliminating tumor cells, but therapy would be chosen for toxicity to quiescent cells. Miller et al 7

The Storage Effect and Multidrug Therapies the environment, which could give differential use of conditions in time by different cancer cell populations; and behaviors The existence of a storage effect implies that additional data are associated with entry into, exit from, and duration of quies- needed to predict tumor progression. Knowing to what extent cence in cancer cells, which could enable buffering of those the variability that one sees in space also occurs in time at each populations to environmental fluctuations, allowing them to spot becomes a valuable and critical piece of information. evade losses in bad times and realize strong growth in good There is recognition of the need for serial histologies, but the times. The possibility that the storage effect is involved in the destructive nature of histological sampling means exact resam- clonal ecological heterogeneity and quiescence/dormancy of pling is impossible. Variable temporal dynamics means that cancer cells, both of which pose serious conundrums for cancer local fluctuations could work against traditional therapies. An therapies, dictates that more attention should be paid to under- underlying storage effect would indicate that cell types differ in standing the temporal dimensions of cancer. To do this, both their dependence on or sensitivity to specific microenviron- the data and the models that inform our understanding of cancer mental conditions that are fluctuating. Each of them could have must be expanded to better reveal and account for temporal an Achilles’ heel. If one can identify or suggest the trade-off variation. If temporal variation in environmental suitability for promoting a storage effect, then a background therapy targeting growth contributes significantly to population interactions quiescent cells could be combined with a standard chemother- within the cancer ecosystem, the implications to how we under- apy and a therapy known to target the specific attributes of the stand and treat cancer would be substantial. cancer cell types. Acknowledgments Caveats The authors thank Drs Sarah Amend, Ryan Bishop, and Chris Whelan for helpful discussions on the content of this article. As a first step, our focus here has been on how the storage effect could contribute toward the diversity of cancer cell types Declaration of Conflicting Interests seen in a tumor. This would be in addition to other sources of intratumor heterogeneity and the many interactions and types The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. of stromal cells such as fibroblasts, macrophages, and peri- cytes. Understanding the full community ecology of these Funding diverse cell types and how the extracellular matrix, for instance, influences therapy resistance is important.68,69 From The author(s) received no financial support for the research, the vantage point of the storage effect, it would be of interest to authorship, and/or publication of this article: AKM and DB were funded by NIH/NCI 1U01CA202958-01. JSB was funded by NIH/ know how these other members of the tumor microenviron- NCI 1U54CA193489-01A1. ment influence temporal and spatial fluctuations in key resources or hazards. If fibroblasts amplify (or dampen) tem- ORCID iD poral variability, then they may facilitate (or prevent) the coexistence of cancer cell types via the storage effect. Anna K. Miller https://orcid.org/0000-0002-3099-1648 Even if the storage effect contributes to 2 coexisting cell types, identifying them poses challenges. First, each should References have a proliferative and quiescent state, meaning that there 1. Warner RR, Chesson PL. Coexistence mediated by recruitment should be 2 identifiably distinct proliferative cell types, 2 dis- fluctuations: a field guide to the storage effect. Am Nat. 1985; tinct quiescent types, and the correct proliferative types must be 125(6):769-787. doi:10.1086/284379 matched with their corresponding quiescent. This is not unlike 2. Huntly N. Herbivores and the dynamics of communities and the challenge of having 2 yeast species in the same sample ecosystems. Annu Rev Ecol Syst. 1991;22(1):477-503. where each yeast species has an active and an encysted form. 3. Chesson P, Huntly N. The roles of harsh and fluctuating condi- Living biopsies or collections of circulating tumor cells may tions in the dynamics of ecological communities. Am Nat. 1997; provide source material for single-cell culturing and the use of 150(5):519-553. doi:10.1086/286080 microfluidics to determine the presence of both life-history 4. Chesson P. Species competition and predation. In: Leemans R, ed. 70,71 stages emerging from the single cell. Ecological Systems. Springer; 2013:223-256. doi:10.1007/978-1- 4614-5755-8_13 5. Amend SR, Pienta KJ. Ecology meets cancer biology: the cancer Summary swamp promotes the lethal cancer phenotype. Oncotarget. 2015; Here, we have described the diversity-promoting mechanism 6(12):9669-9678. doi:10.18632/oncotarget.3430 known as the storage effect and suggested that the conditions 6. Kotler BP, Brown JS. Coexistence mechanisms in the community that make the storage effect relevant to ecosystems in nature ecology of cancer: implications for cancer control. Cancer Con- also are found in cancer. Research is needed to better identify trol. 2020 and quantify the features of cancer that characterize the storage 7. Chesson PL, Huntly N. Community consequences of life-history effect. These features include population-specific responses to traits in a variable environment. Ann Zool Fennici. 1988;25(1):5-16. 8 Cancer Control

8. Chesson P, Huntly N. Short-term instabilities and long-term com- 25. Aguirre-Ghiso JA, Sosa MS. Emerging topics on disseminated munity dynamics. Trends Ecol Evol (Amst). 1989;4(10):293-298. cancer cell dormancy and the paradigm of metastasis. Annu Rev doi:10.1016/0169-5347(89)90024-4 Cancer Biol. 2018;2(1):377-393. doi:10.1146/annurev-cancerbio- 9. Chesson P, Gebauer RLE, Schwinning S, et al. Resource pulses, 030617-050446 species interactions, and diversity maintenance in arid and semi- 26. C´aceres CE. Temporal variation, dormancy, and coexistence: a arid environments. Oecologia. 2004;141(2):236-253. doi:10. field test of the storage effect. Proc Natl Acad Sci U S A. 1997; 1007/s00442-004-1551-1 94(17):9171-9175. doi:10.1073/pnas.94.17.9171 10. Chesson P, Huntly NJ, Roxburgh SH, Pantastico-Caldas M, 27. Chesson P, Huntly N. Temporal hierarchies of variation and the Facelli JM. The storage effect: definition and tests in two plant maintenance of diversity. Plant Species Biol. 1993;8(2-3): communities. In: Kelly CK, Bowler MG, Fox GA, eds. Temporal 195-206. doi:10.1111/j.1442-1984.1993.tb00070.x Dynamics and Ecological Process. Cambridge University Press; 28. Angert AL, Huxman TE, Chesson P, Venable DL. Functional 2013:11-40. doi:10.1017/CBO9781139048170.003 tradeoffs determine species coexistence via the storage effect. 11. Chesson PL, Warner RR. Environmental variability promotes Proc Natl Acad Sci U S A. 2009;106(28):11641-11645. doi:10. coexistence in lottery competitive systems. Am Nat. 1981; 1073/pnas.0904512106 117(6):923-943. doi:10.1086/283778 29. Aguirre-Ghiso JA. Models, mechanisms and clinical evidence for 12. Saxena K, Jolly MK. Acute vs. chronic vs. cyclic hypoxia: their cancer dormancy. Nat Rev Cancer. 2007;7(11):834-846. doi:10. differential dynamics, molecular mechanisms, and effects on 1038/nrc2256 tumor progression. Biomolecules. 2019;9(8)339. doi:10.3390/ 30. Riggi N, Aguet M, Stamenkovic I. Cancer metastasis: a reapprai- biom9080339 sal of its underlying mechanisms and their relevance to treatment. 13. Gillies RJ, Brown JS, Anderson ARA, Gatenby RA. Eco- Annu Rev Pathol. 2018;13:117-140. doi:10.1146/annurev-pathol- evolutionary causes and consequences of temporal changes in 020117-044127 intratumoural blood flow. Nat Rev Cancer. 2018;18(9):576-585. 31. Marusyk A, Polyak K. Tumor heterogeneity: causes and conse- doi:10.1038/s41568-018-0030-7 quences. Biochim Biophys Acta. 2010;1805(1):105-117. doi:10. 14. Michiels C, Tellier C, Feron O. Cycling hypoxia: a key feature of 1016/j.bbcan.2009.11.002 the tumor microenvironment. Biochim Biophys Acta. 2016; 32. Lloyd MC, Cunningham JJ, Bui MM, et al. Darwinian dynamics 1866(1):76-86. doi:10.1016/j.bbcan.2016.06.004 of intratumoral heterogeneity: not solely random mutations but 15. Alve E, Goldstein ST. Dispersal, survival and delayed growth of also variable environmental selection forces. Cancer Res. 2016; benthic foraminiferal propagules. J Sea Res. 2010;63(1):36-51. 76(11):3136-3144. doi:10.1158/0008-5472.CAN-15-2962 doi:10.1016/j.seares.2009.09.003 33. Robertson-Tessi M, Gillies RJ, Gatenby RA, Anderson ARA. 16. Caceres CE. Dormancy in invertebrates. Invertebr Biol. 1997; Impact of metabolic heterogeneity on tumor growth, invasion, 116(4):371-383. doi:10.2307/3226870 and treatment outcomes. Cancer Res. 2015;75(8):1567-1579. 17. Fenelon JC, Renfree MB. The history of the discovery of embryo- doi:10.1158/0008-5472.CAN-14-1428 nic diapause in mammals. Biol Reprod. 2018;99(1):242-251. doi: 34. Ardaˇseva A, Gatenby RA, Anderson ARA, et al. Evolutionary 10.1093/biolre/ioy112 dynamics of competing phenotype-structured populations in per- 18. Lennon JT, Jones SE. Microbial seed banks: the ecological and iodically fluctuating environments. JMathBiol. 2020;80(3): evolutionary implications of dormancy. Nat Rev Microbiol. 2011; 775-807. doi:10.1007/s00285-019-01441-5 9(2):119-130. doi:10.1038/nrmicro2504 35. Vallette FM, Olivier C, L´ezot F, et al. Dormant, quiescent, toler- 19. Baumgartner MF, Tarrant AM. The physiology and ecology of ant and persister cells: four synonyms for the same target in can- diapause in marine copepods. Ann Rev Mar Sci. 2017;9:387-411. cer. Biochem Pharmacol. 2019;162:169-176. doi:10.1016/j.bcp. doi:10.1146/annurev-marine-010816-060505 2018.11.004 20. Tougeron K. Diapause research in insects: historical review and 36. Kwon JS, Everetts NJ, Wang X, et al. Controlling depth of cel- recent work perspectives. Entomol Exp Appl. 2019;167(1):27-36. lular quiescence by an Rb-E2F network switch. Cell Rep. 2017; doi:10.1111/eea.12753 20(13):3223-3235. doi:10.1016/j.celrep.2017.09.007 21. Willis CG, Baskin CC, Baskin JM, et al. The evolution of seed 37. Sosa MS, Bragado P, Aguirre-Ghiso JA. Mechanisms of dissemi- dormancy: environmental cues, evolutionary hubs, and diversifi- nated cancer cell dormancy: an awakening field. Nat Rev Cancer. cation of the seed plants. New Phytol. 2014;203(1):300-309. doi: 2014;14(9):611-622. doi:10.1038/nrc3793 10.1111/nph.12782 38. Meads MB, Gatenby RA, Dalton WS. Environment-mediated 22. De Angelis ML, Francescangeli F, La Torre F, Zeuner A. Stem cell drug resistance: a major contributor to minimal residual disease. plasticity and dormancy in the development of cancer therapy Nat Rev Cancer. 2009;9(9):665-674. doi:10.1038/nrc2714 resistance. Front Oncol. 2019;9:626. doi:10.3389/fonc.2019.00626 39. Wang X, Fujimaki K, Mitchell GC, et al. Exit from quiescence 23. Amend SR, Torga G, Lin K-C, et al. Polyploid giant cancer cells: displays a memory of cell growth and division. Nat Commun. unrecognized actuators of tumorigenesis, metastasis, and resis- 2017;8(1):321. doi:10.1038/s41467-017-00367-0 tance. Prostate. 2019;79(13):1489-1497. doi:10.1002/pros.23877 40. Lawson MA, McDonald MM, Kovacic N, et al. Osteoclasts con- 24. Pienta KJ, Hammarlund EU, Axelrod R, Brown JS, Amend SR. trol reactivation of dormant myeloma cells by remodelling the Poly-aneuploid cancer cells promote evolvability, generating endosteal niche. Nat Commun. 2015;6:8983. doi:10.1038/ lethal cancer. Evol Appl. 2020. doi:10.1111/eva.12929 ncomms9983 Miller et al 9

41. Croucher PI, McDonald MM, Martin TJ. Bone metastasis: the 57. Araujo A, Cook LM, Lynch CC, Basanta D. An integrated com- importance of the neighbourhood. Nat Rev Cancer. 2016;16(6): putational model of the bone microenvironment in bone- 373-386. doi:10.1038/nrc.2016.44 metastatic prostate cancer. Cancer Res. 2014;74(9):2391-2401. 42. Ghajar CM, Peinado H, Mori H, et al. The perivascular niche doi:10.1158/0008-5472.CAN-13-2652 regulates breast tumour dormancy. Nat Cell Biol. 2013;15(7): 58. West JB, Dinh MN, Brown JS, et al. Multidrug cancer therapy in 807-817. doi:10.1038/ncb2767 metastatic castrate-resistant prostate cancer: an evolution-based 43. Lenk L, Pein M, Will O, et al. The hepatic microenvironment strategy. Clin Cancer Res. 2019;25(14):4413-4421. doi:10.1158/ essentially determines tumor cell dormancy and metastatic out- 1078-0432.CCR-19-0006 growth of pancreatic ductal adenocarcinoma. Oncoimmunology. 59. Zhang J, Cunningham JJ, Brown JS, Gatenby RA. Integrating 2017;7(1):e1368603. doi:10.1080/2162402X.2017.1368603 evolutionary dynamics into treatment of metastatic castrate- 44. Archetti M, Pienta KJ. Cooperation among cancer cells: applying resistant prostate cancer. Nat Commun. 2017;8(1):1816. doi:10. game theory to cancer. Nat Rev Cancer. 2019;19(2):110-117. doi: 1038/s41467-017-01968-5 10.1038/s41568-018-0083-7 60. C´ardenas-Navia LI, Mace D, Richardson RA, et al. The pervasive 45. Archetti M. Heterogeneity and proliferation of invasive cancer presence of fluctuating oxygenation in tumors. Cancer Res. 2008; subclones in game theory models of the Warburg effect. Cell 68(14):5812-5819. doi:10.1158/0008-5472.CAN-07-6387 Prolif. 2015;48(2):259-269. doi:10.1111/cpr.12169 61. Turelli M. Niche overlap and invasion of competitors in ran- 46. Gatenby RA, Vincent TL. An evolutionary model of carcinogen- dom environments I. Models without demographic stochasti- esis. Cancer Res. 2003;63(19):6212-6220. city. Theor Popul Biol. 1981;20(1):1-56. doi:10.1016/0040-5809 47. Orlando PA, Gatenby RA, Brown JS. Cancer treatment as a game: (81)90002-2 integrating evolutionary game theory into the optimal control of 62. Chesson P. Multispecies competition in variable environments. chemotherapy. Phys Biol. 2012;9(6):065007. doi:10.1088/1478- Theor Popul Biol. 1994;45(3):227-276. doi:10.1006/tpbi.1994.1013 3975/9/6/065007 63. Dagogo-Jack I, Shaw AT. Tumour heterogeneity and resistance to 48. Basanta D, Hatzikirou H, Deutsch A. Studying the emergence of cancer therapies. Nat Rev Clin Oncol. 2018;15(2):81-94. doi:10. invasiveness in tumours using game theory. Eur Phys J B. 2008; 1038/nrclinonc.2017.166 63(3):393-397. doi:10.1140/epjb/e2008-00249-y 64. Dzobo K, Senthebane DA, Thomford NE, et al. Not everyone fits 49. Kaznatcheev A, Vander Velde R, Scott JG, Basanta D. Cancer the mold: intratumor and intertumor heterogeneity and innovative treatment scheduling and dynamic heterogeneity in social dilem- cancer drug design and development. OMICS. 2018;22(1):17-34. mas of tumour acidity and vasculature. Br J Cancer. 2017;116(6): doi:10.1089/omi.2017.0174 785-792. doi:10.1038/bjc.2017.5 65. Recasens A, Munoz L. Targeting cancer cell dormancy. Trends 50. Basanta D, Gatenby RA, Anderson ARA. Exploiting evolution to Pharmacol Sci. 2019;40(2):128-141. doi:10.1016/j.tips.2018. treat drug resistance: combination therapy and the double bind. 12.004 Mol Pharm. 2012;9(4):914-921. doi:10.1021/mp200458e 66. Napel S, Mu W, Jardim-Perassi BV, Aerts HJWL, Gillies RJ. 51. Basanta D, Scott JG, Fishman MN, et al. Investigating prostate Quantitative imaging of cancer in the postgenomic era: cancer tumour–stroma interactions: clinical and biological radio(geno)mics, deep learning, and habitats. Cancer. 2018; insights from an evolutionary game. Br J Cancer. 2012;106(1): 174-181. doi:10.1038/bjc.2011.517 124(24):4633-4649. doi:10.1002/cncr.31630 52. Anderson ARA. A hybrid mathematical model of solid tumour 67. Jardim-Perassi BV, Huang S, Dominguez-Viqueira W, et al. Mul- invasion: the importance of cell adhesion. Math Med Biol. 2005; tiparametric MRI and coregistered histology identify tumor habi- 22(2):163-186. doi:10.1093/imammb/dqi005 tats in breast cancer mouse models. Cancer Res. 2019;79(15): 53. Gallaher JA, Enriquez-Navas PM, Luddy KA, Gatenby RA, 3952-3964. doi:10.1158/0008-5472.CAN-19-0213 Anderson ARA. Spatial heterogeneity and evolutionary dynamics 68. Senthebane DA, Rowe A, Thomford NE, et al. The role of tumor modulate time to recurrence in continuous and adaptive cancer microenvironment in chemoresistance: to survive, keep your ene- therapies. Cancer Res. 2018;78(8):2127-2139. doi:10.1158/0008- mies closer. Int J Mol Sci. 2017;18(7):1586. doi:10.3390/ 5472.CAN-17-2649 ijms18071586 54. Anderson ARA, Weaver AM, Cummings PT, Quaranta V. Tumor 69. Senthebane DA, Jonker T, Rowe A, et al. The role of tumor morphology and phenotypic evolution driven by selective pres- microenvironment in chemoresistance: 3D extracellular matrices sure from the microenvironment. Cell. 2006;127(5):905-915. doi: as accomplices. IntJMolSci. 2018;19(10):2861. doi:10.3390/ 10.1016/j.cell.2006.09.042 ijms19102861 55. Basanta D, Strand DW, Lukner RB, et al. The role of transforming 70. Sant GR, Knopf KB, Albala DM. Live-single-cell phenotypic growth factor-beta-mediated tumor–stroma interactions in pros- cancer biomarkers-future role in precision oncology? NPJ Precis tate cancer progression: an integrative approach. Cancer Res. Oncol. 2017;1(1):21. doi:10.1038/s41698-017-0025-y 2009;69(17):7111-7120. doi:10.1158/0008-5472.CAN-08-3957 71. Manak MS, Varsanik JS, Hogan BJ, et al. Live-cell phenotypic- 56. Rejniak KA, Anderson ARA. Hybrid models of tumor growth. biomarker microfluidic assay for the risk stratification of cancer Wiley Interdiscip Rev Syst Biol Med. 2011;3(1):115-125. doi:10. patients via machine learning. Nat Biomed Eng. 2018;2(10): 1002/wsbm.102 761-772. doi:10.1038/s41551-018-0285-z